Listen to this manuscript’s

audio summary by

JACC Editor-in-Chief

Dr. Valentin Fuster.

J O U R N A L O F T H E AM E R I C A N C O L L E G E O F C A R D I O L O G Y VO L . 7 0 , N O . 9 , 2 0 1 7

ª 2 0 1 7 B Y T H E AM E R I C A N C O L L E G E O F C A R D I O L O G Y F O U N D A T I O N

P U B L I S H E D B Y E L S E V I E R

I S S N 0 7 3 5 - 1 0 9 7 / $ 3 6 . 0 0

h t t p : / / d x . d o i . o r g / 1 0 . 1 0 1 6 / j . j a c c . 2 0 1 7 . 0 7 . 7 2 3

REVIEW TOPIC OF THE WEEK

Computer-Interpreted ElectrocardiogramsBenefits and Limitations

Jürg Schläpfer, MD,a Hein J. Wellens, MDb

ABSTRACT

Fro

Ins

pa

Ma

Computerized interpretation of the electrocardiogram (CIE) was introduced to improve the correct interpretation of the

electrocardiogram (ECG), facilitating health care decision making and reducing costs. Worldwide, millions of ECGs are

recorded annually, with the majority automatically analyzed, followed by an immediate interpretation. Limitations in the

diagnostic accuracy of CIE were soon recognized and still persist, despite ongoing improvement in ECG algorithms.

Unfortunately, inexperienced physicians ordering the ECG may fail to recognize interpretation mistakes and accept the

automated diagnosis without criticism. Clinical mismanagement may result, with the risk of exposing patients to

useless investigations or potentially dangerous treatment. Consequently, CIE over-reading and confirmation by an

experienced ECG reader are essential and are repeatedly recommended in published reports. Implementation of new ECG

knowledge is also important. The current status of automated ECG interpretation is reviewed, with suggestions for

improvement. (J Am Coll Cardiol 2017;70:1183–92) © 2017 by the American College of Cardiology Foundation.

T he first attempts to automate electrocardio-gram (ECG) analysis go back to the late1950s (1,2), and it was soon expected that

digital computers would have an important role inECG processing and interpretation (3). Despite tech-nical developments, the clinical use of the computer-ized ECG remained initially limited because of thelack of agreement on definitions of waves andcommon measurements, standardized criteria forclassification, and terminology for reporting (4). Toaddress these difficulties, efforts to propose stan-dards and recommendations were developed, bothin Europe and in the United States, to establish aninternational standard for computerized interpreta-tion of the ECG (CIE) (5). The goals were to reducethe wide variation in wave measurements obtainedby ECG computer programs and to assess andimprove the diagnostic classification of ECG interpre-tation (6) so that similar measurements anddiagnostic results could be obtained independent ofthe computer program used (4). However, despite

m the aDepartment of Cardiology, Lausanne University Hospital, Lausan

titute, Maastricht, the Netherlands. The authors have reported that they h

per to disclose.

nuscript received May 30, 2017; revised manuscript received July 5, 2017

all these efforts and advances in the field, an interna-tional accepted standard is still missing (5).

GENERAL COMMENTS ABOUT

TECHNICAL ASPECTS

For digital ECG programs providing diagnostic inter-pretation, several technical aspects have to beconsidered:

1. Signal processing, including acquisition, conver-sion from analog to digital signals, and filtering toeliminate noise (e.g., myopotentials, movementartifacts, baseline wandering linked to respiration).Correct filtering is a fundamental step, as it candramatically alter the final processed signal (5,6).

2. In the majority of automated systems, all ECGleads are now recorded simultaneously. Con-struction of representative template complexes(dominant complexes) excluding premature beatsallows formation of an average complex for eachlead (6).

ne, Switzerland; and the bCardiovascular Research

ave no relationships relevant to the contents of this

, accepted July 11, 2017.

ABBR EV I A T I ON S

AND ACRONYMS

AF = atrial fibrillation

CIE = computerized

interpretation of the

electrocardiogram

ECG = electrocardiogram

LVH = left ventricular

hypertrophy

STEMI = ST-segment elevation

myocardial infarction

Schläpfer and Wellens J A C C V O L . 7 0 , N O . 9 , 2 0 1 7

Computerized Interpretation of the ECG A U G U S T 2 9 , 2 0 1 7 : 1 1 8 3 – 9 2

1184

3. Waveform recognition, with precisedetermination of onset and offset of thedifferent waves (P-wave, QRS complex,T-wave). The temporal alignment andsuperimposition of the representativecomplex for each lead offers more accu-rate labeling of wave onset and offset (6).

4. Measurements of intervals (PR, QRS, QT)and amplitude parameters. When per-formed, global interval measurements areassociated with higher values than single

lead measurements because they remove isoelec-tric intervals present in each of the single leads(7–9). This process is simple and straightforwardwhen the ECG signal is registered in normal sinusrhythm, but it may become very complex in thepresence of atrial arrhythmias (5), requiring time-domain or spectral analysis for recognition anddiscrimination of rapid electrical atrial activity.Manufacturers’ algorithms for determining onsetand offset of waves vary, and are the cause ofrecurrent differences in QRS duration and of dif-ferences in QT interval measurements (10–12).

5. In a recent study, 4 different current digital elec-trocardiographs were studied as to their automatedmeasurement of RR, PR, QRS, and QT intervalduration in 600 ECGs. It included 200 ECGs duringQT interval studies in normal subjects, 200 ECGs innormal subjects during the peak of moxifloxacinadministration (known to modestly prolong the QTinterval), and 200 patients with genotyped variantsof long QT syndrome (8). Measured intervals anddurations show small, but statistically significantgroup differences between manufacturers (8).Mean absolute differences between algorithmswere similar for QRS duration and QT interval innormal subjects, but were significantly larger inpatients with long QT syndrome (8).

Amplitude measurement discrepancies wereless frequently reported, but day-to-day variabilityin amplitude measurements have been described,leading to significant differences in voltagemeasurements and, consequently, in computerdiagnoses (5–10). Despite progress in the develop-ment of the various algorithms, differences inmeasurements results persist, and the call forstandardization and recommendations for defini-tions of waves and references, already initiated inthe 1970s, still remains incompletely answered(10,13). Statements using precise measurementof ECG amplitudes and durations can approachexperienced readers in sensitivity, specificity,and reproducibility (14). However, statements

that depend on waveform configuration (e.g.,repolarization) and relationship between wave-forms (e.g., irregular P waves, atrioventricularconduction disturbances) (Figure 1) may be lessaccurate, as the computer reading the ECG does nothave the visual pattern recognition skills ofa human being (14,15).

6. Interpretation using diagnostic algorithms to theprocessed ECG. These algorithms are proprietary,and may perform differently when applied toECG signals processed by different methods (6).Measurement differences among various standardECG systems may be sufficiently large to alterdiagnostic conclusions (4). This may have clinicalconsequences and, for example, interfere with theselection of candidates for cardiac resynchroniza-tion therapy, as QRS duration is the main deter-minant for device implantation in these patients(11,12).

7. Finally, data compression, transmission, andarchiving are also important aspects of digitalprocessing (5).

ALGORITHM ACCURACY

Algorithm accuracy may vary according to both themanufacturer’s automated program and the level ofthe participating ECGs’ over-readers. Indeed, thesealgorithms are usually tested in comparison with thediagnosis of expert physicians, cardiologists, elec-trophysiologists, or using a consensus of experts (6),considered to be the “gold standard.” Furthermore,ECG interpretation is a mixture of both subjective andobjective aspects, where even experienced cardiolo-gists or experts can disagree, resulting in significantinterobserver variability (16). Additionally, ECGdatabases used in testing computer programs mayinsufficiently represent the overall population; infact, they should be sufficiently large and diverse tocontain all possible clinical diagnoses to mirror dailymedical practice (5). Direct comparative evaluation ofthe performance of commercially available CIE pro-grams has never been performed, mainly due to thereluctance of the manufacturers who own the variousalgorithms. From this perspective, more collaborationamong the various manufacturers would be desirable(Central Illustration).

CURRENT STATUS OF CIE. In 1991, the first system-atic assessment of computer programs compared theperformance of 9 electrocardiographic computerprograms with that of 8 cardiologists in interpretingECGs in 1,220 clinically validated cases of variousdisorders (17). All together, the median total accuracy

FIGURE 1 Atrioventricular Block

Unrecognized 2:1 atrioventricular block. The blocked sinus P-wave is partially hidden in the preceding T-wave.

J A C C V O L . 7 0 , N O . 9 , 2 0 1 7 Schläpfer and WellensA U G U S T 2 9 , 2 0 1 7 : 1 1 8 3 – 9 2 Computerized Interpretation of the ECG

1185

of the computer programs was 6.6% lower than thatof the cardiologists. This comparative study waslimited to a few common ECG diagnostics, such asventricular hypertrophy, old myocardial infarctionsof various locations, and selected combinations ofthese categories (3,17). The performance of someprograms was almost as accurate as the best cardiol-ogists, whereas others were clearly inferior. Further-more, the degree of variability in diagnostic accuracyamong computer programs was significantly greaterthan that among cardiologists (6,17). CIE was found tobe most frequently incorrect in arrhythmias, con-duction disorders, and pacemaker rhythms (18).Arrhythmias . In a study by Shah and Rubin (19),using 2,112 randomly selected, standard 12-lead ECGs,CIE was accurate in interpreting sinus rhythm (posi-tive predictive accuracy 95%), but the performancewas significantly less in nonsinus rhythms (positivepredictive accuracy 53.5%), with the computer unableto produce a rhythm interpretation in 2% of traces.The difficulty in making a correct diagnosis of theunderlying rhythm was linked to recognizing P waveswith a small amplitude (Figure 2), varying P-wavemorphologies or P waves masked by underlying

noise, QRS complexes, or T or U waves (Figure 1) (19).Atrial fibrillation (AF) is another diagnostic challenge.In a recent review Taggar et al. (20) found, whencomparing the automated software diagnosis with theone by health care or primary care professionals, thatCIE showed a borderline greater specificity for AFdiagnosis. Overinterpretation of AF was documentedin 9.3% to 19% of ECGs (21,22), the most prevalentunderlying rhythm in these cases being sinus rhythmand sinus tachycardia with premature atrial beats orbaseline artifacts (Figure 3) (22). Misdiagnosis of AFwas present in 11.3% in the series of Hwan Bae et al.(22), and was more frequent in the elderly. Theserecognized limitations may result in initiation ofunnecessary, potentially harmful medical treatmentor the use of inappropriate diagnostic resources in upto 10% of patients (21,22). Additional failure of theordering physician to correct the false CIE can beanother factor leading to inadequate treatment.Pacemaker rhythms . According to older publica-tions, up to 75% of pacemaker rhythms weremisinterpreted (14,18,19). Failure to recognize low-voltage pulses during pacemaker activity led to mul-tiple errors, such as myocardial infarction in varying

CENTRAL ILLUSTRATION Current Status of Computerized ECG Interpretation andRequired Improvements

COMPUTERIZED INTERPRETATION OF THE ECG (CIE)

Concerns with current CIE: Recommended improvements:

Frequently incorrect readings for:• Arrhythmias• Conduction disorders• Pacemaker rhythms

Wide variations in false-positive and false-negative results

STEMI

Systematic over-readingof CIE is mandatory requiring continuous education and activeECG training

Collaboration among manufacturers in orderto compare and evaluate available algorithms

Standardization of manufacturers’ algorithms

Testing of algorithms using broad ECG databases

Inclusion of age/sex/race in algorithms

in addition to the site of occlusion

Improved cardiologist ECG/CIE training

Schläpfer, J. et al. J Am Coll Cardiol. 2017;70(9):1183–92.

CIE ¼ computerized interpretation of the electrocardiogram; ECG ¼ electrocardiography; STEMI ¼ ST-segment elevation myocardial infarction.

Schläpfer and Wellens J A C C V O L . 7 0 , N O . 9 , 2 0 1 7

Computerized Interpretation of the ECG A U G U S T 2 9 , 2 0 1 7 : 1 1 8 3 – 9 2

1186

locations, left ventricular hypertrophy (LVH), leftbundle branch block, or intraventricular conductiondelay (Figure 4) (18). Substantial progress wasrecently made by many manufacturers now offeringspecific algorithms to allow detection of pacing arti-facts, resulting in significant improvement in pace-maker rhythm diagnosis. However, these algorithmswill need to be adapted to new developments inpacing, such as multisite pacing and resynchroniza-tion therapy.Acute coronary syndromes. The ECG is the mostimportant initial tool in diagnosing an acuteST-segment elevation myocardial infarction (STEMI)(23). ECG misinterpretation and misclassification maydelay or prevent the diagnosis of acute myocardialinfarction, unnecessarily prolonging the door-to-balloon time in the coronary laboratory (24). Auto-mated systems have been developed to diagnoseacute STEMI and tested in the emergency departmentor in the pre-hospital phase to speed up diagnosis and

accelerate early reperfusion. These algorithmsdemonstrate wide variations in false positive (over-diagnosis in 0% to 42%) and false negative results(underdiagnosis in 22% to 42%) (25). Those discrep-ancies were illustrated by Garvey et al. (26), whorecently showed the varying accuracy of 3 differentavailable STEMI diagnostic algorithms to identify thelocation of the culprit coronary artery lesion. In thesestudies, different ECG machines with various algo-rithms were tested in patient groups with a differentprevalence of STEMI. Also, CIE diagnoses werecompared with interpretations from heterogeneoussources: cardiologists, emergency physicians, WorldHealth Organization criteria, discharge diagnosis ofSTEMI, or catheterization laboratory findings (25).Because of its high false negative results in theidentification of STEMI, it is not recommended thatCIE be used as the sole means to activate the cardiaccatheterization laboratory. It should always be usedin conjunction with physicians, paramedics, or nurses

FIGURE 2 Unrecognized Sinus Rhythm

Unrecognized regular sinus rhythm with double counting of the heart rate (135 beats/min) due to T-wave oversensing. Consequently, the QTc

interval is overestimated at 509 ms; note that no QT prolongation warning appears in the diagnosis report.

J A C C V O L . 7 0 , N O . 9 , 2 0 1 7 Schläpfer and WellensA U G U S T 2 9 , 2 0 1 7 : 1 1 8 3 – 9 2 Computerized Interpretation of the ECG

1187

trained to recognize STEMI (25). Preferably, the ECGreview should be supplemented by the characteristicsof the patient’s symptoms, the past medical history,and comparison with previous ECGs to improve CIEdiagnostic accuracy and help in the decision process(Central Illustration) (27).

Bosson et al. (27) recently showed that ECGartifacts and nonischemic causes of ST-segmentelevation, such as early repolarization, were themost common reasons for incorrect algorithm inter-pretation of STEMI. Minimizing ECG artifacts bytraining paramedics on how to recognize and avoidthem is critical in improving the performance ofsoftware (27).

For the future, electrocardiograph manufacturersare asked to equip their machines with switchingsystems allowing the arrangement of limb leads intheir anatomically contiguous sequence (23), whichmay help to make the correct diagnosis. Algorithmsshould be further refined to display the spatialorientation of the ST-segment deviation vector in thefrontal and transverse planes, and to help in locatingthe culprit vessel as well as the site of occlusion (23).

These algorithms should also help clinicians torecognize, in acute coronary settings, culprit lesionsites resulting in a large ischemic area that requirerapid reperfusion, as in left main obstruction, prox-imal left anterior descending artery occlusion, andmultivessel disease, or in situations in which the ECGinterpretation is complicated by conduction disorders(e.g., complete left bundle branch block or pacedrhythms) (28). Recording additional right-sidedprecordial leads should also be suggested in infero-posterior infarctions to make the distinctionbetween a right or circumflex coronary artery occlu-sion (6,23). Incorporating sex, age, race, and priorECG findings into algorithms could also increase thesensitivity of a STEMI diagnosis (Central Illustration)(23–25).Repolar i zat ion . QT interval. Precise QT intervalmeasurement is another challenge for computerizedalgorithms, as the reliability of QT measurements haslong been considered to be limited. Miller et al. (29)showed that diagnostic accuracy of screening ECGs inlong QT syndrome was unsatisfactory and failed toidentify at-risk family members. Significant progress

FIGURE 3 Misdiagnosis of Atrial Fibrillation

Normal sinus rhythm recorded during a checkup. The poor quality of the electrocardiogram made the fellow trust the wrong automated

diagnosis of atrial fibrillation (AF) despite readily visible P waves. bpm ¼ beats per minute.

Schläpfer and Wellens J A C C V O L . 7 0 , N O . 9 , 2 0 1 7

Computerized Interpretation of the ECG A U G U S T 2 9 , 2 0 1 7 : 1 1 8 3 – 9 2

1188

has been made since then, with newer algorithms forQT measurements in good-quality tracings, but mea-surement errors remain in abnormal or poor-qualityECGs (30). The automatically measured global QTinterval is generally longer than the QT intervalmeasured in any individual lead by as much as 30 to40 ms (9–31). As previously described, individual leadmeasurements tend to underestimate the true QTinterval, as the onset or offset of waveforms may beisoelectric in a specific lead, whereas the lead with theearliest QRS onset may be different from that with thelatest T-wave end (3,7,31). Superimposing individualtracings or points used to derive the QT interval shouldhelp the physician to obtain the correct value (3). A QTinterval may be within or outside of the acceptednormal values depending on the measurementmethod (visual or automatic) used. Central databaseswith high-quality ECGs using comparable methodol-ogy will be the only way to distinguish normal fromabnormal values in specific populations (31). Differ-ences in QT measurements between algorithms fromthe same manufacturer may exist, stressing theimportance of using an electrocardiograph equipped

with the same algorithm when measuring QT intervalvalues during ECG surveillance in patients with longQT syndrome treated with drugs potentially affectingthe QT interval (32,33). The same rule should beapplied during serial comparisons in drug testing or forregulatory authorities in their evaluation of a possibleproarrhythmic risk of a newdrug (32,33). Recently Gargand Lehmann (34) drew attention to the fact that thealgorithmic diagnosis did not display a prolonged QTinterval in 52.5% of patients with a prolonged QT in-terval. This was attributed to ECG waveform–basedcriteria included in the tested algorithm (34). A similarbehavior could occur (but not yet be published) withthe algorithms of other manufacturers (Figure 2). Theunder-reporting of prolonged QTc interval in patientson methadone was also recognized (33). Before defin-itively concluding that the displayed QTc interval iscorrect, both systematic checking of the actual QTcvalue and visual validation of QT interval prolongationare strongly recommended (9,32–34).Early repolarization. The early repolarization patternis a widely prevalent condition, lately linked with anincreased risk of arrhythmic death (35). Recent

FIGURE 4 Unrecognized Pacing Activity

Failure to recognize both atrial flutter and ventricular pacing activity. AV ¼ atrioventricular.

J A C C V O L . 7 0 , N O . 9 , 2 0 1 7 Schläpfer and WellensA U G U S T 2 9 , 2 0 1 7 : 1 1 8 3 – 9 2 Computerized Interpretation of the ECG

1189

development of sensitive algorithms may allowautomatic detection of early repolarization in thegeneral population, which could provide useful in-formation for research programs or future large reg-istries established to more precisely defineindividuals at risk of dying suddenly (35–38).

Left ventr i cu lar hypertrophy . The use of echo-cardiography, magnetic resonance imaging, orcomputerized tomography questioned the value ofthe ECG in diagnosing LVH. However, its lower costand ease of use still justify its place in daily practice,epidemiological studies, and clinical trials (39). Overthe years, different, initially simple electrocardio-graphic criteria were used for the diagnosis of LVH.Recently, more complex criteria, based on products ofvoltage and QRS duration, QRS area, or composite useof several criteria, have been published. However, nosingle diagnostic criterion can be recommended (39)due to their low sensitivity and specificity. CIE mayhelp ECG readers save time in applying all criteriavalidated for identifying LVH and adjusted fornoncardiac factors such as age, sex, race, and body

habitus. The final interpretation should specify whichdiagnostic criteria were used and which wereabnormal (39) (Central Illustration).CIE as a screen ing tool . Berge et al. (40) showed, in595 professional soccer players, that abnormal ECGswere more than twice as common after computer-based measurements than after visual measure-ments. The investigators appropriately suggest thatthis may justify redefining or adjusting referencevalues for abnormalities, as previously done forautomatic QRS width and QT interval measurements(10–12,31). These observations raise the more generalquestion of the influence of the technique used forECG interpretations (visual diagnostic or automatedmeasurements), not only in athletes in whom ECGscreening value is controversial, but also in otherpopulations (31). Homogeneity in ECG interpretationtechniques is mandatory to ensure coherent results inresearch or epidemiological studies using ECGscreening (40–42).

ARTIFACTS. Artifacts are well known to lower ECGdiagnostic accuracy (Figure 3). To allow algorithms to

Schläpfer and Wellens J A C C V O L . 7 0 , N O . 9 , 2 0 1 7

Computerized Interpretation of the ECG A U G U S T 2 9 , 2 0 1 7 : 1 1 8 3 – 9 2

1190

perform at their best, quality recording is essential.Nurses and technicians in charge of ECG recordingsshould be well taught about the importance of correctpreparation of the patient by cleaning the skin andusing electrodes with contact gel. Periodic retrainingin proper lead positioning of the different leadsshould also be routinely organized. Next, algorithmsshould alert for improper filter use and for anyinterchange of limb lead connections or precordialelectrode misplacements (Central Illustration) (6).Paramedics involved in ECG recordings should also beeducated to immediately alert the physician or thebedside nurse when a critical diagnosis is given by themachine (15).

PRACTICAL ASPECTS. Provided that the ECG qualityis good, computerized interpretation gives a correctmeasurement of the basic parameters. Computer-assisted ECG interpretation decreased analysis timeby up to 24% to 28% for experienced readers (5). Also,computerized archives allow rapid access to serialECG comparisons. Besides indicating differencesbetween ECGs, it improves interpretation accuracy,for example, in acute coronary syndromes (43).

CIE has the goal of improving patient care bydisplaying information helpful to reducing medicalerrors, facilitating physicians’ reading, and reducingcosts (44,45). However, computer-based analysis ofthe ECG may lead to erroneous diagnosis with useless,inappropriate, or even dangerous care of the patient.These potentially harmful consequences stress therequirement for continuous development in softwareand systematic over-reading of the ECG (5,6).

The correct interpretation of an ECG remains achallenge for physicians with a low level of knowl-edge (Figure 3) (45). Internal medicine residents havea low overall proficiency and self-perceived confi-dence in interpreting ECGs. They also find theirtraining insufficient (45–47). Cardiologists as primaryreaders more often correct the misinterpreted ECGs,as compared with internists or others specialists (48).In the United States, cardiology fellows are requestedto interpret approximately 3,000 to 3,500 ECGs dur-ing their standard 3-year training program to acquirecompetence in ECG interpretation (49). Training toreview, edit, and amend ECGs generated by thecomputerized system that provides preliminaryinterpretation is part of their training (49).

The consensus is to consider CIE only as an adjunctto, and not a substitute for, the electrocardiographer.Consequently, all computer-based reports should besystematically over-read (5,6). It requires training ofphysicians and health personnel to validate (or not)the automated interpretation and early detection of

errors (49,50). Knowledge of the strengths andlimitations of CIE is a prerequisite for avoiding blindtrust in software interpretations (5). Automatedinterpretation may influence physicians’ ECG reading.It improves their diagnostic abilities when the inter-pretation is correct, but increases the probability oferrors when the proposed diagnosis is incorrect, andnegatively influences the decision making of thephysician in charge of the patient (44,50,51). It hasbeen roughly estimated that these misdiagnoses mayaccount for up to 10,000 adverse effects or avoidabledeaths worldwide annually (16). Additionally, level ofexpertise, training of the physician, time constraints,or fatigue are other main causes of diagnostic errors inECG interpretation (21,48,52).

No single method of teaching appears most effec-tive in obtaining ECG interpretation skills (53,54).However, repeated assessments, tested by a finalexamination, increase medium-term retention of ECGinterpretation skills, whatever the instruction format.Continuous ECG training, coupled with appropriateexaminations should also be done at the post-graduate level (55).

Standardization in applied algorithms and uni-formization of diagnostic criteria and statementshave been proposed (56). It would lead to worldwideuniformity of ECG interpretation and facilitate thedevelopment of a uniform teaching curriculum inelectrocardiography.

The ECG is a noninvasive, powerful, and cost-effective tool when interpreted by the properspecialist, but hospitals may have difficulties infinding the resources to offer the ECG standard of careand comply with advised guidelines (16). A formalprocess of over-reading of all ECGs by cardiologists,or ECG-trained and ECG-tested physicians or para-medics, should be the rule and routinely applied inhospitals and clinics (5,16). Ideally, a clear deliveryand routing system has to be organized, with over-reading of the ECG and rapid transmission to thepatient’s chart (16). When not possible, an alternativeapproach would be collaboration with an outside,centralized ECG interpretation service (16).

CONCLUSIONS

Significant progress was made in the development ofECG algorithms for use in the CIE. However, limita-tions are still present, requiring standardization, withcontinuous improvement in applied software anduniformization of ECG diagnostic criteria and state-ments. Systematic over-reading of CIE is mandatory.To provide patients with the best standard of care,critical knowledge in ECG interpretation remains

J A C C V O L . 7 0 , N O . 9 , 2 0 1 7 Schläpfer and WellensA U G U S T 2 9 , 2 0 1 7 : 1 1 8 3 – 9 2 Computerized Interpretation of the ECG

1191

necessary, and can only be acquired by continuouseducation and active ECG training. It also requires theimplementation of new ECG information. Closecooperation between clinical ECG experts and CIEmanufacturers is needed to optimize CIE performance.

ADDRESS FOR CORRESPONDENCE: Dr. Jürg Schläpfer,Department of Cardiology, Lausanne University Hospital,Rue du Bugnon 44, 1011 Lausanne, Switzerland. E-mail:jurg.schlaepfer@chuv.ch.

RE F E RENCE S

1. Taback L, Marden E, Mason HL, Pipberger HV.Digital recording of electrocardiographic data foranalysis by a digital computer. IRE Trans MedElectro 1959;6:167–71.

2. Pipberger HV, Freis ED, Taback L, Mason HL.Preparation of electrocardiographic data for anal-ysis by digital electronic computer. Circulation1960;21:413–8.

3. Rautaharju PM. Eyewitness to history: land-marks in the development of computerized elec-trocardiography. J Electrocardiol 2016;49:1–6.

4. Willems JL, Arnaud P, van Bemmel JH, et al.Assessment of the performance of electrocardio-graphic computer programs with the use of areference data base. Circulation 1985;71:523–34.

5. Hongo RH, Goldschlager N. Status of comput-erized electrocardiography. Cardiol Clin 2006;24:491–504.

6. Kligfield P, Gettes LS, Bailey JJ, et al. Recom-mendations for the standardization and interpre-tation of the electrocardiogram. Part I: theelectrocardiogram and its technology. A scientificstatement from the American Heart AssociationElectrocardiography and Arrhythmias Committee,Council on Clinical Cardiology; the AmericanCollege of Cardiology Foundation; and the HeartRhythm Society endorsed by the InternationalSociety for Computerized Electrocardiology. J AmColl Cardiol 2007;49:1109–27.

7. Kligfield P, Tyl B, Mareek M, Maison-Blanche P.Magnitude, mechanism, and reproducibility of QTinterval differences between superimposed globaland individual lead ECG complexes. Ann Nonin-vasive Electrocardiol 2007;12:145–52.

8. Kligfield P, Badilini F, Rowlandson I, et al.Comparison of automated measurements ofelectrocardiographic intervals and durations bycomputer-based algorithms of digital electrocar-diographs. Am Heart J 2014;167:150–9.e1.

9. Rautaharju PM, Surawicz B, Gettes LS, et al.AHA/ACCF/HRS recommendations for the stan-dardization and interpretation of the electrocar-diogram: part IV: the ST segment, T and U waves,and the QT interval: a scientific statement fromthe American Heart Association Electrocardiog-raphy and Arrhythmias Committee, Council onClinical Cardiology; the American College ofCardiology Foundation; and the Heart RhythmSociety. Endorsed by the International Society forComputerized Electrocardiology. J Am Coll Cardiol2009;53:982–91.

10. Willems JL. A plea for common standards incomputer aided ECG analysis. Comput Biomed Res1980;13:120–31.

11. Van�cura V, Wichterle D, Ul�c I, et al. The vari-ability of automated QRS duration measurement.Europace 2017;19:636–43.

12. De Pooter J, El Haddad M, Stroobandt R, DeBuyzere M, Timmermans F. Accuracy of computer-calculated and manual QRS duration assessments:clinical implications to select candidates for car-diac resynchronization therapy. Int J Cardiol 2017;236:276–82.

13. Recommendations for measurement standardsin quantitative electrocardiography. The CSEWorking Party. Eur Heart J 1985;6:815–25.

14. Poon K, Okin PM, Kligfield P. Diagnosticperformance of a computer-based ECG rhythmalgorithm. J Electrocardiol 2005;38:235–8.

15. Alpert JS. Can you trust a computer to readyour electrocardiogram? Am J Med 2012;125:525–6.

16. Mele PF. The ECG dilemma: guidelines onimproving interpretation. J Healthc Risk Manag2008;28:27–31.

17. Willems JL, Abreu-Lima C, Arnaud P, et al. Thediagnostic performance of computer programs forthe interpretation of electrocardiograms. N Engl JMed 1991;325:1767–73.

18. Guglin ME, Thatai D. Common errors incomputer electrocardiogram interpretation. Int JCardiol 2006;106:232–7.

19. Shah AP, Rubin SA. Errors in computerizedelectrocardiogram interpretation of cardiacrhythm. J Electrocardiol 2007;40:385–90.

20. Taggar JS, Coleman T, Lewis S, Heneghan C,Jones M. Accuracy of methods for diagnosingatrial fibrillation using 12-lead ECG: a systematicreview and meta-analysis. Int J Cardiol 2015;184:175–83.

21. Bogun F, Anh D, Kalahasty G, et al. Misdiag-nosis of atrial fibrillation and its clinical conse-quences. Am J Med 2004;117:636–42.

22. Hwan Bae M, Hoon Lee J, Heon Yang D, et al.Erroneous computer electrocardiogram interpre-tation of atrial fibrillation and its clinical conse-quences. Clin Cardiol 2012;35:348–53.

23. Wagner GS, Macfarlane P, Wellens H, et al.AHA/ACCF/HRS recommendations for the stan-dardization and interpretation of the electrocar-diogram: part VI: acute ischemia/infarction: ascientific statement from the American HeartAssociation Electrocardiography and ArrhythmiasCommittee, Council on Clinical Cardiology; theAmerican College of Cardiology Foundation; andthe Heart Rhythm Society. Endorsed by theInternational Society for Computerized Electro-cardiology. J Am Coll Cardiol 2009;53:1003–11.

24. Hollander JE, Than M, Mueller C. State-of-the-art evaluation of emergency departmentpatients presenting with potential acute coronarysyndromes. Circulation 2016;134:547–64.

25. O’Connor RE, Al Ali AS, Brady WJ, et al. Part 9:Acute coronary syndromes: 2015 American HeartAssociation guidelines update for cardiopulmonaryresuscitation and emergency cardiovascular care.Circulation 2015;132:S483–500.

26. Garvey JL, Zegre-Hemsey J, Gregg R,Studnek JR. Electrocardiographic diagnosis of STsegment elevation myocardial infarction: anevaluation of three automated interpretationalgorithms. J Electrocardiol 2016;49:728–32.

27. Bosson N, Sanko S, Stickney RE, et al. Causesof prehospital misinterpretations of ST elevationmyocardial infarction. Prehosp Emerg Care 2017;21:283–90.

28. Meissner A, Trappe HJ, de Boer MJ,Gorgels AP, Wellens HJ. The value of the ECG fordecision making at first medical contact in thepatient with acute chest pain. Netherlands Heart J2010;18:301–6.

29. Miller MD, Porter Cb, Ackerman MJ. Diagnosisaccuracy of screening electrocardiograms in longQT syndrome I. Pediatrics 2001;108:8–12.

30. Tyl B, Azzam S, Blanco N, Wheeler W.Improvement and limitation of the reliability ofautomated QT measurement by recent algorithms.J Electrocardiol 2011;44:320–5.

31. Maron BJ, Friedman RA, Kligfield P, et al.,American Heart Association Council on ClinicalCardiology, Advocacy Coordinating Committee,Council on Cardiovascular Disease in the Young,Council on Cardiovascular Surgery and Anesthesia,Council on Epidemiology and Prevention, Councilon Functional Genomics and Translational Biology,Council on Quality of Care and OutcomesResearch, and American College of Cardiology.Assessment of the 12-lead ECG as a screening testfor detection of cardiovascular disease in healthygeneral populations of young people (12–25 yearsof age): a scientific statement from the AmericanHeart Association and the American College ofCardiology. J Am Coll Cardiol 2014;64:1479–514.

32. Estes NA 3rd. Computerized interpretation ofECGs: supplement not a substitute. Circ ArrhythmElectrophysiol 2013;6:2–4.

33. Talebi S, Azhir A, Zuber S, et al. Under-estimated and unreported prolonged QTc byautomated ECG analysis in patients on methadone:can we rely on computer reading? Acta Cardiol2015;70:211–6.

34. Garg A, Lehmann MH. Prolonged QT intervaldiagnosis suppression by a widely used comput-erized ECG analysis system. Circ Arrhythm Elec-trophysiol 2013;6:76–83.

35. Macfarlane PW, Antzelevitch C,Haissaguerre M, et al. The early repolarizationpattern: a consensus paper. J Am Coll Cardiol2015;66:470–7.

Schläpfer and Wellens J A C C V O L . 7 0 , N O . 9 , 2 0 1 7

Computerized Interpretation of the ECG A U G U S T 2 9 , 2 0 1 7 : 1 1 8 3 – 9 2

1192

36. Clark EN, Katibi I, Macfarlane PW. Automaticdetection of end QRS notching or slurring.J Electrocardiol 2014;47:151–4.

37. Kenttä T, Porthan K, Tikkanen JT, et al.Sensitivity and specificity of automated detectionof early repolarization in standard 12-lead elec-trocardiography. Ann Noninvasive Electrocardiol2015;20:355–61.

38. Patton KK, Ellinor PT, Ezekowitz M, et al.,American Heart Association Electrocardiographyand Arrhythmias Committee of the Council onClinical Cardiology and Council on FunctionalGenomics and Translational Biology. Electrocar-diographic early repolarization: a scientific state-ment from the American Heart Association.Circulation 2016;133:1520–9.

39. Hancock EW, Deal BJ, Mirvis DM, et al. AHA/ACCF/HRS recommendations for the standardiza-tion and interpretation of the electrocardiogram:part V: electrocardiogram changes associated withcardiac chamber hypertrophy: a scientificstatement from the American Heart AssociationElectrocardiography and Arrhythmias Committee,Council on Clinical Cardiology; the AmericanCollege of Cardiology Foundation; and the HeartRhythm Society: endorsed by the InternationalSociety for Computerized Electrocardiology. J AmColl Cardiol 2009;53:992–1002.

40. Berge HM, Steine K, Andersen TE, Solberg EE,Gjesdal K. Visual or computer-based measure-ments: important for interpretation of athletes’ECG. Br J Sports Med 2014;48:761–7.

41. Dubin AM. Screening ECGs for youngcompetitive athletes: it is complicated. Curr OpinPediatr 2015;27:604–8.

42. Drezner JA, O’Connor FG, Harmon KG, et al.AMSSM position statement on cardiovascularpreparticipation screening in athletes: currentevidence, knowledge gaps, recommendations andfuture directions. Br J Sports Med 2017;51:153–67.

43. Gregg RE, Deluca DC, Chien CH,Helfenbein ED, Ariet M. Automated serial ECGcomparison improves computerized interpretationof 12-lead ECG. J Electrocardiol 2012;45:561–5.

44. Tsai TL, Fridsma DB, Gatti G. Computer deci-sion support as a source of interpretation error:the case of electrocardiograms. J Am Med InformAssoc 2003;10:478–83.

45. Salerno SM, Alguire PC, Waxman HS. Compe-tency in interpretation of 12-lead electrocardio-grams: a summary and appraisal of publishedevidence. Ann Intern Med 2003;138:751–60.

46. Goy JJ, Schlaepfer J, Stauffer JC. Competencyin interpretation of the 12-lead electrocardiogramamong Swiss doctors. Swiss Med Wkly 2013;143:w13806.

47. Eslava D, Dhillon S, Berger J, Homel P,Bergmann S. Interpretation of electrocardiogramsby first-year residents: the need for change.J Electrocardiol 2009;42:693–7.

48. Anh D, Krishnan S, Bogun F. Accuracy ofelectrocardiogram interpretation by cardiologistsin the setting of incorrect computer analysis.J Electrocardiol 2006;39:343–5.

49. Balady GH, Bufalino VJ, Gulati M, Kuvin JT,Mendes LA, Schuller JL. COCATS 4 Task Force 3:training in electrocardiography, ambulatory elec-trocardiography and exercise testing. J Am CollCardiol 2015;65:1763–77.

50. Martínez-Losas P, Higueras J, Gómez-Polo JC,et al. The influence of computerized interpretationof an electrocardiogram reading. Am J Emerg Med2016;34:2031–2.

51. Southern WN, Arnsten JH. The effect of erro-neous computer interpretation of ECGs on resi-dent decision making. Med Decis Making 2009;29:372–6.

52. Spodick DH. Computer error and acceptablestandards for electrocardiographic interpretation.J Electrocardiol 2006;39:348–9.

53. Rowlands D, Moore P. ECG interpretation inthe NHS. Br J Cardiol 2014;21:47–8.

54. Fent G, Gosai J, Purva M. Teaching the inter-pretation of electrocardiograms: which method isbest? J Electrocardiol 2015;48:190–3.

55. Raupach T, Harendza S, Anders S, Schuelper N,Brown J. How can we improve teaching of ECGinterpretation skills? Findings from a prospectiverandomized trial. J Electrocardiol 2016;49:7–12.

56. Mason JW, Hancock EW, Gettes LS, et al.Recommendations for the standardization andinterpretation of the electrocardiogram: part II:electrocardiography diagnostic statement list ascientific statement from the American HeartAssociation Electrocardiography and ArrhythmiasCommittee, Council on Clinical Cardiology; theAmerican College of Cardiology Foundation; andthe Heart Rhythm Society endorsed by the Inter-national Society for Computerized Electro-cardiology. J Am Coll Cardiol 2007;49:1128–35.

KEY WORDS algorithms, software